Rakuten

We consider a multi-way massive multi-input multi-output (mMIMO) relaying system wherein a relay aids multi-way data exchange between multiple users by employing non-orthogonal multiple access (NOMA). The relay achieves this by superposing different user signals in the power domain. Each user then sequentially decodes the data of all other users by performing successive interference cancellation (SIC). To perform SIC, a user utilizes the downlink channel information, which it estimates using the precoded pilots transmitted by the relay. We derive a closed-form spectral efficiency (SE) expression for this downlink-pilot-aided NOMA multi-way relaying system by considering spatially-correlated channels, and imperfect SIC at the user. We next design two algorithms to maximize this SE by optimally allocating the user transmit powers and the NOMA variables. Both these algorithms provide a similar SE, but the latter one, due to its closed-form power updates, has a much lesser complexity. The efficacy of downlink pilots in multi-way NOMA relaying, with practical correlated Rayleigh-faded channels, is shown by demonstrating that they vastly outperform the case when users perform SIC using channel statistics.

Image restoration in adverse weather conditions is a difficult task in computer vision. In this paper, we propose a novel transformer-based framework called GridFormer which serves as a backbone for image restoration under adverse weather conditions. GridFormer is designed in a grid structure using a residual dense transformer block, and it introduces two core designs. First, it uses an enhanced attention mechanism in the transformer layer. The mechanism includes stages of the sampler and compact self-attention to improve efficiency, and a local enhancement stage to strengthen local information. Second, we introduce a residual dense transformer block (RDTB) as the final GridFormer layer. This design further improves the network’s ability to learn effective features from both preceding and current local features. The GridFormer framework achieves state-of-the-art results on five diverse image restoration tasks in adverse weather conditions, including image deraining, dehazing, deraining & dehazing, desnowing, and multi-weather restoration. The source code and pre-trained models will be released.

Most video anomaly detection approaches are based on non-semantic features, which are not interpretable, and prevent the identification of anomaly causes. Therefore, we propose a caption-guided interpretable video anomaly detection framework that explains the prediction results based on video captions (semantic). It utilizes non-semantic features to fit the dataset and semantic features to provide common sense and interpretability to the model. It automatically stores representative anomaly prototypes and uses them to guide the model based on similarity with these prototypes. Specifically, we use video memory to represent the content of videos, which includes video features (non-semantic) and caption information (semantic). The proposed method generates and updates a memory space during training, and predicts anomaly scores based on the memory similarities between the input video and the stored memories. Furthermore, the stored captions can be used as descriptions of representative anomaly actions. Moreover, the proposed module can be easily integrated with any existing method by replacing a linear layer, owing to its strong usability. The interpretability and reliable detection performance of the proposed method are empirically evaluated on popular benchmark datasets and analyzed through extensive experiments.

The classification of gender from handwriting is a challenging issue that have great attention recently. Most of the exiting works were conducted on gender classification using face image and off-line handwritten texts. This study explored an automated system for gender classification from online handwritten patterns. The handwritten samples were collected from 79 (Male: 32 and Female: 47) using pen tablet device. Each subject was asked to perform four tasks such as Zigzag trace (ZigZ-T), Zigzag predict (ZigZ-P), periodic line trace (PL-T), and periodic line predict (PL-P) and repeated its into three times. Thirty-six statistical features were derived from the six raw features, obtained from Pen-tablet device. Following that, we selected the best subset of features by employing Sequential Forward Floating Selection (SFFS)-based algorithm. At the same time, four machine learning (ML)-based algorithms like support vector machine (SVM), random forest, AdaBoost, and Gradient Boosting (GB) were employed for gender classification.We trained these four ML-based algorithms with leave-one-out method and optimized their hyperparameters. The experimental results showed that SVM achieved a recognition accuracy of 88.10% for adult ZigZ-T tasks and 90.09% recognition accuracy was obtained by GB-based algorithm for children whose drawing ZigZ-P tasks. our proposed system demonstrates promise in automating gender classification based on handwriting patterns, offering insights into the significant differences between adult and child handwriting. The ability to identify gender accurately from handwriting has broad applications, including security enhancements and personalized service provision.

Variational Bayesian learning (VBL)-based sparse channel state information (CSI) estimation is conceived for multiple input multiple output (MIMO) orthogonal time frequency space (OTFS) and for orthogonal time sequence multiplexing (OTSM)-based systems relying on low-resolution analog-to-digital convertors (ADCs). First, the CSI estimation model is developed for MIMO-OTFS systems considering quantized outputs. Then a novel VBL technique is developed for exploiting the inherent DD domain sparsity. Subsequently, an end-to-end system model is derived for MIMO-OTSM systems, once again, using only finite-resolution ADCs. Similar to OTFS systems, it is demonstrated that the channel is sparse in the delay-sequency (DS)-domain. Thus the sparse CSI estimation problem of the MIMO-OTSM system can also be solved using the VBL technique developed for its OTFS counterpart. A bespoke minimum mean square error (MMSE) receiver is developed for data detection, which unlike the conventional MMSE receiver also accounts for the quantization error. Finally, finite-resolution ADCs emerge as a solution, offering reduced costs and energy consumption amid the growing challenge posed by energy-intensive high-resolution ADCs in Next-Generation (NG) systems. The efficacy of the proposed techniques is validated by simulation results, surpassing the state-of-the-art and signalling a transition towards more sustainable communication technologies.

Interconnected systems forming complex networks are ubiquitous in many man-made and natural phenomena. When individual systems are aligned towards a desired trajectory, their synchronization’s stability depends on the network’s controllability, often achieved through pinning control. When optimizing controllability, unweighted Laplacian and uniform feedback gains are conventionally used for the pinned nodes, ignoring the significance of link weights, usually leading to suboptimal results. This study improves local stability of the synchronous state by addressing the controllability problem using weighted Laplacian matrices and nonuniform feedback gains. Using the master-stability function method, direct optimization of the controllability measure is formulated as a multi-objective optimization problem with a Pareto frontier and multiple optimal points. This multi-objective optimization problem is simplified into a spectral radius minimization problem, where reformulating it as a semidefinite programming (SDP) problem has led to a unique optimal point on its Pareto frontier. Many interesting analytical results have been established for different families of networks and subnetworks, including the criteria for optimal zero weights that can divide the network or optimal controllability measure of an arbitrary network and its mirrored networks. Additionally, for deterministic scale-free networks, it is demonstrated how the network can break down into smaller replicas and ultimately form a collection of path networks. Numerical simulations using the Rössler model illustrate the feasible region and, interestingly, show that the Pareto frontier of the deterministic scale-free networks is independent of its size.

Efficiently running federated learning (FL) on resource-constrained devices is challenging since they are required to train computationally intensive deep neural networks (DNN) independently. DNN partitioning-based FL (DPFL) has been proposed as one mechanism to accelerate training where the layers of a DNN (or computation) are offloaded from the device to the server. However, this creates significant communication overheads since the intermediate activation and gradient need to be transferred between the device and the server during training. While current research reduces the communication introduced by DNN partitioning using local loss-based methods, we demonstrate that these methods are ineffective in improving the overall efficiency (communication overhead and training speed) of a DPFL system. This is because they suffer from accuracy degradation and ignore the communication costs incurred when transferring the activation from the device to the server. This article proposes Eco Fed-a communication efficient framework for DPFL systems. Eco Fed-a eliminates the transmission of the gradient by developing pre-trained initialization of the DNN model on the device for the first time. This reduces the accuracy degradation seen in local loss-based methods. In addition, EcoFed proposes a novel replay buffer mechanism and implements a quantization-based compression technique to reduce the transmission of the activation. It is experimentally demonstrated that EcoFed can reduce the communication cost by up to 133× and accelerate training by up to 21× when compared to classic FL. Compared to vanilla DPFL, EcoFed achieves a 16× communication reduction and 2.86× training time speed-up. EcoFed is available from https://github.com/blessonvar/EcoFed .

Recent developments in deep learning have greatly improved the accuracy of various natural language processing tasks. Patent classification using patent documents has also seen a significant improvement in accuracy based on the internationally defined IPC as well as the region-specific CPC prediction methods. Such an automated patent classification capability can reduce the burden on examiners and improve efficiency. It can also be used to organize patents, such as in prior art searches. However, such a classification cannot be used to support the formulation of management strategies, such as capturing the fields in which the predicted patent classification will be used in the management of a company. To this end, this study aims to classify patents based on technological field instead of the existing IPC classification such that the developed automated patent classification is useful for companies. To elaborate, this study investigates the differences between IPC and the proposed classification based on technological field, selects documents to be used, and proposes a classification model using IPC information. The proposed classification model is comprised of two distinct components, namely a model for document input and another for IPC input. BERT is employed for the document input model, while skip-connections are used for the IPC input model. Finally, the improvement in accuracy is examined compared to existing IPC using actual data; moreover, solutions are proposed to the problems identified in this study. As a result, our proposed model significantly improves precision, recall, F1, and AP over existing models.

As the number of articles on postsecondary topics expands, new methods are required to quantitatively understand the literature. Previous scholars looking at the higher education literature use manual coding, which limits the number of years that can be studied, or network analysis of citations and words, which does not yield groupings of articles by topic area. Instead, we use topic modeling to understand the subject areas that scholars investigate, as well as changes in these subject areas over time. Topic modeling assumes that a group of abstracts contains a mix of topics that are hidden (or latent) because we can only observe abstracts and the words that appear within abstracts, but not the underlying topics. Each abstract and word are then viewed as having a probability of belonging to a topic or subject area. Our data consist of abstracts from the set of articles published in The Journal of Higher Education, Research in Higher Education, and Review of Higher Education between 1991 and 2020. We find 24 main topics in the postsecondary literature in the past three decades. The most common topics in the literature during the past three decades are research usage and research methodology (18%), followed by college access (9%), identities and experiences (9%), student engagement (9%), and academic careers (8%). The research topics that became more popular over time are all student related: identities and experiences, college access, financial aid, student experiences with diversity, and student success. Topics that became less popular over time include academic misconduct, research usage and research methodology, and academic careers.

The networked Susceptible-Infected-Susceptible (SIS) model is a well-studied model for epidemic spreading processes in complex networks. A critical problem in the networked SIS model with time delay is to simultaneously optimize the network's convergence rate to the healthy state and its robustness to time delay while there is a constraint on the total curing resources that can be utilized throughout the network. Here, this problem has been derived as a multi-objective optimization problem with additional constraints guaranteeing a minimum convergence rate and a maximum tolerable time delay. It is shown that the problem's Pareto frontier consists of three segments, where for complete bipartite network and uniform infection rates their details have been derived analytically. By incorporating both objective functions into one, a corresponding convex optimization problem is developed, which is solved over tree reducible bipartite networks, using semidefinite programming (SDP). Furthermore, a convex combination approximation has been provided for the second segment of the Pareto frontier. The trade-off between the convergence rate and robustness to time-delay has been demonstrated via numerical simulations, where, for scale-free networks, the impact of network's size on this trade-off has been investigated numerically. Additionally, it is shown how the convergence rate is dictated by the eigenvalue of the M -matrix of the network, and the first segment of the Pareto frontier provides the most diverse results.